Quality and Operations
Management
Process Control and Capability Analysis
Process Control
Recognizes that variance exists in all processes
Sources of variation
systematic
assignable
Purpose
to detect and eliminate out-of-control conditions
to return a process to an in-control state
Basic tool -- the SPC chart(s)
Measuring A Process
Types of measurements
variables data
length, weight, speed, output, etc
discrete values
attributes data
good vs bad, pass vs fail, etc
binary values
Types of charts
variables -- X-R chart
attributes -- p, np, c and u
Basic assumption -- sample means are normally
distributed
Getting Started with SPC
X-R Charts
Determine sample size and frequency of data collection
Collect sufficient historical data
Ensure normality of distribution
Calculate factors for control charts
X
UCLx
LCLx
Construct
R control
LCLr chart UCLr
Plot data points
Determine outliers and eliminate assignable causes
Recalculate control limits with reduced data set
Implement new process control chart
Basic Properties
x = std dev of sample mean = /n
(where = process standard deviation)
conventional approach uses 3 /n
limitations of control charts
Type I Error: probability that an in-control value would appear as
out-of-control
Type II Error: probability that a shift causing an out-of-control
situation would be mis-reported as in-control
delays due to sampling interval
charting without taking action on assignable causes
over control actions
Type 1 and Type 2 Error
In Control
Out of Control
Alarm
No Alarm
Type 1
error
No error
No error
Type 2
error
Suppose 1 > , then
Type 2 Error = Z [( + 3 x - 1) / x ]
Type 1 Error = 0.0027 for 3 charts
Type 2 Error Example
Suppose:
= 10
1= 10.2
= 4/3
n=9
thus,
x = 4/9
Then, Type 2 Error
= Z [( + 3 x - 1) / x ]
= Z [(10 + 12/9 - 10.2) / (4/9)]
= Z [2.55] = 0.9946
if 1= 11.0, then Type 2 Error
= Z[0.75] = .7734
if 1= 12.0, then Type 2 Error
= Z[-1.50] = .0668
Prob.{shift will be detected in 3rd sample after shift occurs}
= 0.0668*0.0668*(1-0.0668) = 0.0042
Average number of samples taken before shift is detected
= 1/(1-0.0668) = 1.0716
Prob.{no false alarms first 32 runs, but false alarm on 33rd}
= (0.9973)32*(0.0027) = .0025
Average number of samples taken before a false alarm
= 1/0.0027 = 370
Tests for Unnatural Patterns
Probability that odd patterns observed are not
natural variability are calculated by using the
probabilities associated with each zone of the
control chart
Use the assumption that the population is normally
distributed
Probabilities for X-chart are shown on next slide
Normal Distribution Applied to
X-R Control Charts
Probability = .00135
UCLx
Outer 3rd
Probability = .02135
+3
Middle 3rd
Probability = .1360
+2
Inner 3rd
B
C
Probability = .3413
+1
Inner 3rd
Probability = .3413
Middle 3rd
Probability = .1360
-2
Outer 3rd
Probability = .02135
-3
Probability = .00135
LCLx
-1
A Few Standard Tests
1 point outside Zone A
2 out of 3 in Zone A or above (below)
4 out of 5 in Zone B or above (below)
8 in a row in Zone C or above (below)
10 out of 11 on one side of center
Tests for Unnatural Patterns
2 out of 3 in A or beyond
.0227 x .0227 x (1-.0227) x 3 = .0015
4 out of 5 in B or beyond
.15874 x (1-.1587) x 5 = .0027
8 in a row on one side of center
.508 = .0039
Other Charts
P-chart
based on fraction (percentage) of defective units in a
varying sample size
np-chart
based on number of defective units in a fixed sample size
u-chart
based on the counts of defects in a varying sample size
c-chart
based on the count of defects found in a fixed sample size
SPC Quick Reference Card
P-chart
based on fraction (percentage) of defective units in a varying sample size
UCL/LCLp = p 3(p)(1-p)/n
np-chart
based on number of defective units in a fixed sample size
UCL/LCLnp = np 3(np)(1-p)
u-chart
based on the counts of defects in a varying sample size
UCL/LCLu = u 3u/n
c-chart
based on the count of defects found in a fixed sample size
UCL/LCLc = c 3c
X-R chart
variables data
UCL/LCLX = X 3 x = X 3 / n = X A2R where R/d2
UCLR = D4R
and A2 = 3/d2 n
LCLR = D3R
for p, np, u, c and R chart the LCL can not be less than zero.
Process Capability
Cp: process capability ratio
a measure of how the distribution compares to the width of the
specification
not a measure of conformance
a measure of capability, if distribution center were to match
center of specification range
Cpk: process capability index
a measure of conformance (capability) to specification
biased towards worst case
compares sample mean to nearest spec. against distribution
width
How Good is Good Enough?
Cp = 1.0 => 3 => 99.73% (in acceptance) => .9973 => 2700
ppm out of tolerance
PG&E operates non-stop
23.65 hours per year without electricity
average car driven 15,000 miles per year
41 breakdowns or problems per year
Cp = 2.0 => 6 => 99.99983% (in acceptance) => .9999983
=> 3.4 ppm out of tolerance
PG&E operates non-stop
1.78 minutes per year without electricity
average car driven 15,000 miles per year
0.051 breakdowns or problems per year or one every 20 years
